Senthilnathan K

Founder @Sudotank - Building Euler, the Data Engine for Physical AI.

[Senthilnathan K @senthsei]

1/ Euler's technical whitepaper is live. excited to finally share this.

The claim is simple: robot data should not reach training just because it exists. It should prove it is ready.

We measured Euler on real teleoperation data, controlled faults, and ground truth.

[Senthilnathan K @senthsei]

@petergyang They turn on fire mode just when they are about to lose, what a team!

[Senthilnathan K @senthsei]

Nice share, one takeaway here is that the bottleneck isn’t Claude’s reasoning, it’s the control interface. Once you let the model supervise a strong pretrained policy instead of wrestling raw joints, performance jumps dramatically. It’s the classic tools > raw-intelligence moment.

[Senthilnathan K @senthsei]

@adnothing This is massive!
Russ’s lessons in robotics has had the most foundational impact personally, great to see this team up.

[Senthilnathan K @senthsei]

@gowthami_s The interesting challenges around simulators, and reward design def makes it fun. What’s the most enjoyable part for you so far? I’d be interested in any key lessons or tools you’ve found valuable.

[Senthilnathan K @senthsei]

Absolutely, and this is a huge opportunity to be that middleware that leverages the logistical prowess of vendors for collection, and can also can work with robot labs to actually steer the captured data towards more usability and training-readiness.

[Chris Paxton @chris_j_paxton]

I dont think you can collect great data without understanding robot learning

[Senthilnathan K @senthsei]

@ShreyaR It’s about time vendors adopt a forward-deployed model of sorts to actually know what kind of data to collect and render it usable from the client’s perspective, rather than just playing the logistics game and scaling up data-capture nobody wants.

[Senthilnathan K @senthsei]

@LeRobotHF How does this compare to using a frontier VLM like Gemini 3.5 straight up, is there any published benchmark I can refer to?

[Senthilnathan K @senthsei]

@OfficialLoganK Can’t agree more and that’s what we are building, Euler does it for modalities beyond language and to solve the huge bottleneck in the Physical AI niche

[Senthilnathan K @senthsei]

@_varunnair This is great work, Varun, happy to chat more about this, will DM.

[Senthilnathan K @senthsei]

@lockedincheetah Physical world disruption’s coming, it’s time we software heads start learning to catch up

[Senthilnathan K @senthsei]

@brandonguo There’s no rigorous QA process standards for data capture, which clearly makes it a low-hanging fruit for founders to grab onto.
Making existing data pipelines reliable and ready for training and optimising that for model performance is the real deal.

[Senthilnathan K @senthsei]

@BrendanFoody True, nowadays companies are moving beyond evals to having their own custom benchmarks to thrive

[Senthilnathan K @senthsei]

@livinoffwater That’s insane!

[Senthilnathan K @senthsei]

@livinoffwater How intense is the robotics/Physical-AI scene there when compared to other clusters like SF/Boston? Been wanting to come witness what’s unfolding in Shenzhen

[Senthilnathan K @senthsei]

@oprydai Does it count if you make the software that makes it easier to build hardware

[Senthilnathan K @senthsei]

@MilaidyNow Ironically, the goal however is to automate that same compensated labour and outsource it to embodiments isn’t it

[Senthilnathan K @senthsei]

@shivanijpatel My gradient descent won’t chase any global optimum if that’s the case

[Senthilnathan K @senthsei]

@marcklingen Boredom is the best form of mindfulness, like the Dutch practice - Niksen

[Senthilnathan K @senthsei]

@bryan_johnson Embrace mortality